KR20090108036A - Quality Scalable Video Data Stream - Google Patents

Abstract

An apparatus for generating a quality-scalable video data stream (36) is described which comprises means (42) for coding a video signal (18) using block-wise transformation to obtain transform blocks (146, 148) of transformation coefficient values for a picture (140) of the video signal, a predetermined scan order (154, 156, 164, 166) with possible scan positions being defined among the transformation coefficient values within the transform blocks so that in each transform block, for each possible scan position, at least one of the transformation coefficient values within the respective transform block belongs to the respective possible scan position; and means (44) for forming, for each of a plurality of quality layers, a video sub-data stream (30; 28, 30) containing scan range information indicating a sub-set of the possible scan positions, and transform coefficient information on transformation coefficient values belonging to the sub-set of possible scan positions such that the sub-set of each quality layer comprises at least one possible scan position not comprised by the sub-set of any other of the plurality of quality layers.

Description

Quality scalable video data stream

The present invention relates to a picture quality-scalable video data stream, wherein video data stream generation and decoding is obtained using block-wise conversion.

The present invention relates to a picture quality-scalable video data stream, wherein video data stream generation and decoding is obtained using block-wise conversion.

The current joint video team "JVT" of the ITU-T Video Coding Expert Group from the ISO / IEC Video Expert Group (MPEG) lists the scalable extension of the current H.264 / MPEG4-AVC video coding standard. A key feature of scalable video coding (SVC) compared to conventional single layer coding is that various representations of video sources having different resolutions, frame rates and / or bit rates are provided within a single bit stream. Video representations with a particular spatial-temporal resolution and bit rate can be extracted from the global SVC bit-stream by simple stream manipulation such as packet dropping. As an important feature of the SVC design, most configurations of H.264 / MPEG4-AVC are used as specified in the standard. This includes motion-compensation and internal prediction, transform and entropy coding, and deblocking as well as NAL unit packetization. The base layer of an SVC bit-stream is generally coded according to H.264 / MPEG4-AVC, so each standard that is suitable for an H.264 / MPEG4-AVC decoder will have a base layer representation when it is provided as an SVC bit-stream. Can be decoded. New tools are just being added to support spatial and SNR scalability.

For SNR scalability, coarse-grain / medium-grain scalability (CGS / MGS) and fine-grain scalability (GFS) are distinguished in the current draft specification. Coarse-grain or medium-grain SNR scalable coding is performed by using a concept similar to that for spatial scalability. Pictures of other SNR layers are independently coded with layer specific motion parameters. However, in order to improve the coding efficiency of enhanced layers as compared to simulcast, additional inter-layer prediction mechanisms have been introduced. These prediction mechanisms have been made so that the encoder can freely determine and change whether the base layer information will be used for efficient enhancement layer coding. Since the embodied inter-layer prediction concepts include techniques for motion parameters and residual prediction, the temporal prediction structure of SNR layers will be temporarily adjusted for efficient use of inter-layer prediction. It is known that all NAL units turn into excess units for a moment and then follow the SVC bit-stream inside each other. The following three inter-layer prediction techniques are included in the SVC design.

The first is called inter-layer motion prediction. In order to use base-layer motion data for enhancement layer coding, additional macroblock modes have been introduced in SNR enhancement layers. Macroblock partitioning is obtained by copying the partitioning of co-located macroblocks in the base layer. Combined motion vectors as well as reference picture indices are copied from the co-located base layer blocks. In addition, the motion vector of the base layer can be used as a motion vector predictor for the conventional macroblock mode.

The second technique of redundancy reduction between the various picture quality layers is called inter-layer residual prediction. The use of inter-layer residual prediction is indicated by the flag (residual_prediction_flag) sent for all inter-coded macroblocks. When this flag is true, the base layer signal of the co-located block is used as a prediction for the residual signal of the current macroblock, so that only the corresponding difference signal is coded.

Finally, inter-layer internal prediction is used to take advantage of redundancy between layers. In this inner-macroblock mode, the prediction signal is established by the co-located reconstruction signal of the base layer. For inter-layer internal prediction, it is generally required that the base layers be fully decoded, including the computationally complex operation of motion-compensated prediction and deblocking. However, it has been known that this problem can be avoided when inter-layer internal prediction is limited to portions of an intra-coded lower layer image. With this limitation, each supported target layer can be decoded with a single motion compensation loop. This single-loop decoding mode is essential for scalable H.264-MPEG4-AVC extension.

Since inter-layer inner prediction can only be applied when co-located macroblocks are inner-coded, and inter-layer motion prediction including macroblock type guessing can only be applied when base-layer macroblocks are inter-coded, Both modes are represented via a single syntax element base_mode_flag on the macroblock level. When this flag is equal to 1, inter-layer inner prediction is determined when the base layer macroblock is inner-coded. On the other hand, macroblock mode as well as motion vectors and reference indices are copied from the base layer macroblock.

In order to support finer granularity than CGS / MGS coding, so-called progressive fine slices have been introduced to enable fine SNR scalable coding (FGS). Each progressive fine slice represents a refinement of the residual signal corresponding to a halving of the quantization step size (QP increase of 6). These signals are only represented in such a way that a single inverse transform is performed for each transform block at the decoder side. The order of the transform coefficient levels in progressive fine slices causes the corresponding NAL units to be computed at any arbitrary byte-arrayed point so that the quality of the SNR base layer is refined in a fine-granular way. In addition to refinement of the residual signal, it may also transmit refinement of the motion parameters as part of the progressive fine slices.

One drawback of FGS coding in the current SVC draft is that it obviously increases decoder complexity compared to CGS / MGS coding. In one aspect the transform coefficients in the progressive fine slice are coded using various scans for the transform blocks, and only a small number of transform coefficient levels are transmitted in each scan. For the decoder, high memory bandwidth is required, which increases complexity, since all transform coefficient levels from other scans need to be collected before the inverse transform is performed. On the other hand, the parsing process for progressive fine slices depends on the syntax elements of the corresponding base layer slices. The possible model selection for the order or arithmetic coding of syntax elements as well as codeword tables for VLC coding depends on the syntax elements in the base layer. Since the syntax elements of the base layer must be accessed during the parsing of the enhancement layer, it further increases the required memory bandwidth for decoding.

In addition, the special nature of progressive fine slices that can be truncated is difficult to use in today's packet switch networks. Typically, a media aware network device will send or drop a packet of scalable bit-screens. And the only error that can occur at the application layer is packet loss.

Therefore, it is desirable to have a coding scheme that is more suitable for today's needs, with other video compression techniques as well as the aspect of H.264-MPEG4-AVC, and showing fewer packets rather than byte truncation problems.

According to one embodiment of the present invention, an image quality-scalable video data stream generating apparatus, in each transform block, for each possible scan position, at least one of the transform coefficient values in each transform block is each possible scan position. A preset scan order with the possible scan positions is defined between transform coefficient values in a transform block, and uses block-wise transform to obtain transform blocks of transform coefficient values for an image of the video signal. Means for encoding a video signal to perform; And a transform coefficient for the transform coefficient values belonging to the sub-set of the possible scan positions such that the sub-set of each quality layer includes at least one possible scan position not covered by the other multiple sub-sets of quality layers. Means for forming, for each of the plurality of quality layers, a video sub-data stream comprising information and scan area information indicative of the sub-set of the possible scan positions.

Furthermore, according to one embodiment of the present invention, an apparatus for reconstructing a video signal from a picture quality-scalable video data stream comprising a video sub-data stream for each of a plurality of picture quality layers is provided in each transform block. For each possible scan position, at least one of the transform coefficient values in each transform block is defined between the transform coefficient values in the transform blocks so that a preset scan order with the possible scan positions falls within the respective possible scan position. And for each image quality layer, to obtain transform coefficient information and scan region information for transform coefficient values of different transform blocks, and the scan region information indicating a subset of the possible scan positions. Means for parsing video sub-data streams of the camera; Using the scan region information, for each of the picture quality layers, to build the transform block by combining the transform coefficient values of the respective transform blocks from the transform coefficient information into a subset of the possible scan positions. Way; And means for reconstructing the image of the video signal by inverse transform of the transform blocks.

1 shows a block diagram of an encoder for generating a picture quality-scalable video data stream according to one embodiment of the invention.

2 illustrates a block diagram of the high-layer hybrid coder of FIG. 1 in accordance with one embodiment of the present invention.

3 illustrates a block diagram of the base-layer hybrid coder of FIG. 1 in accordance with one embodiment of the present invention.

4 illustrates a block diagram of a layer coding unit of the high quality layer of FIG. 1 according to an embodiment of the present invention.

FIG. 5 shows a schematic diagram showing the structure of an image as well as block transform according to an embodiment of the present invention.

6A-6G show schematic diagrams of the division of the transform block into sub-layers and the scanned portion in accordance with various embodiments of the invention.

7 shows a schematic diagram illustrating the construction of sub-data streams according to one embodiment of the invention.

8 shows pseudo-code illustrating the coding of transform coefficient levels belonging to a particular sub-data stream in accordance with one embodiment of the present invention.

9 shows a pseudo-code illustrating another embodiment for encoding transform coefficient levels belonging to a particular sub-data stream.

10 shows a block diagram of a decoder according to another embodiment of the present invention.

11 is a block diagram of an embodiment of the decoding unit of FIG. 10.

Based on the drawings, preferred embodiments of the present invention are described in more detail below.

1 shows an encoder for quality-scalable bit-stream generation. For example, the encoder 10 shown in FIG. 1 is dedicated to generate a scalable bit-stream capable of supporting two different spatial layers and (N + 1) signal-to-noise ratio (SNR) layers. For this purpose, the encoder 10 consists of a base layer part 12 and a spatial enhancement layer part 14. The image quality reduction unit 16 of the encoder 10 receives an original or higher definition video 18 representing a series of images 20, and receives an original video 18 composed of a series of images 24. In order to obtain the low quality version 22 and the low quality version 22 input to the base layer part 12, its image quality-meaning spatial resolution in the example of FIG. 1-is reduced. For example, the image quality reduction unit 16 may perform subsampling of images using two subsampling factors. However, although FIG. 1 illustrates an example of supporting two spatial layers 12 and 14, the embodiment shown in FIG. 1 reduces the image quality performed between the original video 18 and the low quality video 22. Should be readily applicable to applications that do not include sub-sampling (e.g., reducing the representation bit-depth of a pixel value or when a picture quality reduction unit simply copies an input signal to an output signal).

While the base layer part 12 receives the low quality video 22, the original video 18 is input to the high quality part 14, and the two parts 12 and 14 respectively perform hybrid coding for the input video. To perform. The base layer part 12 receives the low quality video 22 and generates a base layer bit-stream 26. High-quality layer part 14, on the other hand, receives original video 18 and, in addition to spatial enhancement layer bit-stream 28, N signal-to-noise ratio (SNR) refinement layer bit-streams. SNR refinement layer bit-streams 30 are generated. The correlation and generation of the spatial enhancement layer bit-stream 28 and the base layer bit-stream 26 will be described in more detail below. As a precaution only, base layer part 12 may add some signal-to-noise ratio (SNR) refinement layer bit-streams 32 to base layer bit-stream 26. However, in order to more easily explain the laws of the embodiment according to the present invention, it is assumed that SNR scalability is limited to the enhancement layer part 14. However, the discussion below will show that the correlation regarding the high quality layer part 14 and the SNR refinement layers described below can be easily applied to the base layer part 12. This refers to the dashed line 32 of FIG. 1.

As described in more detail below, all bit-streams 26 to 32 are input to a multiplexer 34 which generates a scalable bit-stream 36 from data streams at the input arranged in packets. .

Internally, the base layer part 12 has a layer code unit 38 and a hybrid coder 38 connected in series between the low quality video 24 input on one side and the multiplexer 34 on the other side as described above. unit 40). Similarly, the high quality layer part 14 may include a hybrid coder 42 and a layer coding unit 44 connected between the original video 18 input and the multiplexer 44. Each hybrid coder 42, 38, respectively, uses hybrid coding (i.e., motion compensated prediction used with block-wise transformation of the prediction residual). Code the input signal that is input to it. Thus, each hybrid coder 42, 38 has a respective layer coding unit in which each motion information data 46, 48 is serialized together with the respective residual data 50, 52. Output to the inputs of 40 and 44, respectively.

Naturally, redundancy exists between the motion data 46 on one side and the motion data 48 on the other side and the residual data 52 on the other side. This inter-layer redundancy is utilized by the hybrid coder 42. More specifically, based on the macroblock basis, hybrid coder 42 can select one of several interlayer prediction options. For example, hybrid coder 42 may decide to use or adopt base layer motion data 46 as motion data 48 for the high quality layer. Optionally, the hybrid coder 42 may decide to use the base layer motion data 46 as a predictor of the motion data 48. As another option, the hybrid coder 42 may code the motion data 48 as completely new, i.e., independent of the base layer motion data.

Similarly, hybrid coder 42 may code the residual data 42 as prediction residual corresponding to base layer residual data 50 as a predictor to use for prediction for the high quality layer. However, the hybrid coder 42 may also in this case be a predictor for the picture content of the original video, so that the motion data and / or the residual data 48, 52, respectively, can simply code the residual associated with the reconstructed base layer data. Reconstruction of the image content of the base layer may also be used. As will be described with reference to FIG. 2, the reconstructed base layer image information is dedicated reconstructed connected between the base layer hybrid coder 38 or the base-layer coding unit 40 and the high quality layer hybrid coder 42. It may be received from a reconstruction unit 54.

Subsequently, the internal structures and functions of the base layer hybrid coder 38 and the high quality layer hybrid coder 42 will be described in detail with the layer coding unit 44. With respect to the layer coding unit 40, the following assumes that the layer coding unit 40 simply generates the base layer data-stream 26. However, an optional embodiment for layer coding unit 40 that generates SNR refinement layer data-streams 32 as described above is also described in layer coding unit 44 described below. It can be easily deduced from the related description.

First, the internal structure and function of the base layer hybrid coder 38 are described. As shown in FIG. 3, the base layer hybrid coder 38 includes an input 56 for receiving a low quality video signal, an output 58 for motion data 46, and an output 60 for residual data 50. ), An output 62 for connecting the motion data 58 to the hybrid coder 42, an output 64 for connecting the reconstructed base layer image data to the hybrid coder 42, and a residual data 50. It may include an output 66 for coupling to the hybrid coder 42.

Internally, the hybrid coder 38 has a transform unit 68, a back-transformation unit 70, a subtractor 72, an adder 74, and a motion prediction unit. (motion prediction unit) 76. As already mentioned, the subtractor 72 and the conversion unit 68 are connected between the input 56 and the output 60. Subtractor 72 subtracts the motion-predicted video content received from motion prediction unit 76 from the input video signal, and sends a difference signal to transform unit 68. . Transform unit 68 performs block-wise transformation on the difference / residue signal and optionally performs quantization on the transform coefficients. The conversion result is output to the output 60 by the conversion unit 68 and also to the input of the inverse conversion unit 70. The inverse transform unit 70 eventually performs inverse transform on the transform block of the transform coefficient according to the quantization progress. The result is the reconstructed residual signal in which the predicted motion video content output from the motion prediction unit 76 is additionally combined by the adder 74. The addition result performed by the adder 74 is a reconstructed video of base quality. The output of the adder 74 is output to the output 64 and simultaneously to the input of the motion prediction unit 76. Motion prediction unit 76 performs motion-compensated prediction on the reconstructed images to predict other images of the video input via input 56. While motion-prediction is performed, motion prediction unit 76 generates motion data that includes motion vectors, motion image, etc. that reference indices, and outputs this mode motion data to output 58. Output to the output 62. The output of the transform unit 68 is also connected to the output 66 for delivering transform residual data to the hybrid coder 42 of the high quality layer. As already mentioned above, the functions of the two hybrid coders 38 and 42 shown in FIG. 1 are similar to each other. Thus, the structure of the hybrid coder 42 shown in FIG. 2 is similar to that of the hybrid coder 38 shown in FIG. More specifically, the hybrid coder 42 includes an input 86 for the original video signal 18, an output 88 for motion data 48, an output 90 for residual data 52, and a base layer hybrid. It may include three inputs 92, 94, 96 for coupling to the output 62, output 64, and output 66 of the coder 38, respectively. Internally, hybrid coder 42 may include two switches or selectors 98, 100 for connecting one of two paths 102, 104 between input 86 and output 90. Can be. More specifically, path 104 is a subtractor 106, transform unit 108 and residual predictive coder 110 connected between input 86 and output 90 via switches 98 and 100. ) May be included. Subtractor 106 and transform unit 108 are combined by inverse transform unit 112, adder 114, and motion prediction unit 116 by components 68, 76 of hybrid coder 38 shown in FIG. 3. Form a prediction loop as it is formed. Thus, the transformed version of the motion-predicted residual result at the output of transform unit 108 is input to residual prediction coder 110. Residual prediction coder 110 is also coupled to input 96 to receive base layer residual data. By using the base layer residual data as a predictor, the residual prediction coder 110 codes a portion of the residual data output from the transform unit 108 as the prediction residual associated with the residual data of the input 96. do. For example, the residual prediction coder 100 up-samples the base layer residual data and subtracts the upsampled residual data from the residual data output from the transform unit 108. Of course, the residual prediction coder 110 may simply perform prediction of a portion of the residual data output from the transform unit 108. The other paths pass through the remaining predictive coder 110 unchanged. The granularity of these parts can be macro blocks. In other words, a determination regarding residual data at input 96 may be used as a predictor or may be indicated by a macroblock basis and a separate syntax element residual_prediction_flag. It may not be performed on the outcome of the decision.

Similarly, hybrid coder 42 is a motion parameter predictive coder for receiving motion data output from base layer and motion information obtained from motion prediction unit 116 via input 92. coder) 118, the motion parameter prediction coder 118 passes to the output 88 without changing the motion data output from the motion prediction unit 116 based on the macroblock basis, or input 92 The motion data is predictively coded using motion information output as a predictor from the base layer. For example, motion parameter prediction coder 118 may code the motion vectors obtained from motion prediction unit 116 as offset vectors associated with the motion vectors included in the base layer motion data at input 92. Optionally, the motion parameter prediction coder 118 forwards the base layer information output from the input 92 to the motion prediction unit 116 for use in motion prediction of the high quality layer. In this case, there is no motion data to be transmitted for each part of the high quality layer video signal. As another option, the motion parameter prediction coder 118 ignores the presence of motion data at the input 92 and codes the motion data from the motion prediction unit 116 directly to the output 88. Determination of these possibilities is coded into the resulting picture quality scalability bit-stream.

Finally, predictive coder 120 is provided at path 102 and coupled to input 94. Prediction coder 120 predicts portions of the high definition layer based on each portion of the reconstructed base layer video signal so that each residual or difference can simply be advanced at the output of prediction coder 120. Prediction coder 120 also operates on a macroblock-based basis that cooperates with switches 98 and 100.

As shown in FIG. 4, the layer coding unit 44 of the high quality layer receives motion data from the input 122 and the output 88 for receiving transform coefficients of the residual data from the output 90. It may include an input 124 for receiving. Distributing unit 126 receives the transform coefficients and distributes the received transform coefficients to several enhancement layers. The distributed transform coefficients are therefore output to the formation unit 128. Together with the distributed transform coefficients, forming unit 128 receives motion data from input 124. The forming unit 128 combines both data, and based on such data inputs, refinement-layer data streams 30 and zero-order enhancement layer data streams. form a stream 28.

In order to enable a more detailed description of the functions of the distribution unit 126 and the forming unit 128, the following is performed by the conversion and distribution unit 126 performed by the block-based conversion unit 108. Its internal relationship to the distribution to be described will be described in more detail with reference to FIG. 5 shows an image 140. For example, image 140 is part of high definition video data 18. For example, the pixels in the image 140 are arranged horizontally and horizontally. For example, the image 140 is divided into macroblocks 142 regularly arranged vertically and horizontally. For example, each macroblock 142 spatially covers a square image area to contain 16x16 samples of luma components of the image. More specifically, the macroblocks 142 can be organized in pairs of macroblocks. More specifically, two vertically adjacent macroblocks 142 form such a pair of macroblocks and may spatially have a macroblock pair region 144 of the image 140. Through macroblock pair basis, the hybrid coder 42 (FIG. 1) allows macroblocks 142 in individual regions 144 in field-mode or frame-mode. ) Can be processed. In the case of field-mode, video 18 is assumed to include two inserted fields (top field and bottom field), where the top field includes an even numbered rows of pixels, The bottom field includes odd numbered rows starting at the second line of the image 140. In this case, the top macroblock of region 144 is related to the pixel values of the top field lines in region 144, while the bottom macroblock of region 144 is related to the content of the remaining lines. Thus, in this case, both macroblocks have a reduced vertical resolution and occupy all regions of the region 144 in sufficient space. In the case of frame-mode, the top macroblock is defined to surround the top half of the columns within the region 144, while the bottom macroblock contains residual image samples in the region 144.

As already mentioned above, the conversion unit 108 performs block-wise conversion of the residual signal output from the subtractor 106. In this regard, the block basis for transform in transform unit 108 may be different from the macroblock size of macroblocks 142. More specifically, each macroblock 142 may be divided into four (ie, 2x2) transform blocks 146 or 16 (ie, 4x4) transform blocks 148. Using the example for the macroblock size of the same 16x16 image samples, the transform unit 108 blocks-blocks the macroblocks 142 of the image 140 into blocks of the size of 4x4 pixel samples or 8x8 pixel samples. To convert. Accordingly, the transform unit 108 may have several transform blocks 146, 148, i.e., sixteen 4x4 transform coefficient blocks or four 8x8 transform coefficient blocks, for a particular macroblock 142. outputs transform coefficient blocks).

At 150, an example of an 8x8 transform coefficient block of a frame-coded macroblock is shown. More specifically, at 150, each transform coefficient is assigned a scan position number 0-63, which is indicated by this. As represented by axes 152, each transform coefficient is associated with a different spatial frequency component. More specifically, the frequency associated with one of the transform coefficients increases in magnitude from the upper left corner of the transform block 150 to the lower right corner. The scan order defined by the scan position among the transform coefficients of the transform block 150 scans the transform coefficients in a zigzag manner from the top left to the bottom right, and this zigzag scan is shown by arrow 154.

Just for completeness, a scan between transform coefficients may be defined differently than between transform coefficients of a transform block for a field-coded macroblock. For example, for field-coded macroblocks, as shown at 156 in FIG. 5, the transform coefficient scan 158 is in the reciprocating or zigzag direction from the top left to the bottom right (frame at 150). The transform coefficients are scanned in a zigzag fashion with a steep zigzag direction than the 45 degree zigzag direction used in the case of coded macroblocks. More specifically, coefficient scan 158 performs vertical transform coefficients twice as fast as the transverse direction to account for the fact that field-coded macroblocks surround the image samples with a vertical pitch twice the parallel or transverse pitch. Scan Thus, as in the case of coefficient scan 154, coefficient scan 158 scans the transform coefficients in such a way that the frequency may increase as the position scan number increases.

Examples of coefficient scans for 8x8 transform coefficient blocks are shown at 150 and 158. However, as described above, there may also be transform blocks of smaller size (ie, 4x4 transform coefficients). In this case, respective position scans are shown at 160 and 162 of FIG. 5, while scan 164 of 160 is dedicated to frame-coded macroblocks, while scan of 162. 166 is dedicated for field-coded macroblocks.

Specific embodiments related to the size and arrangement of the macroblocks and transform blocks shown in FIG. 5 are merely exemplary, and various transforms and applications may be possible. Before beginning the detailed description of the series of features, image 140 may be subdivided into several slices 168 based on the macroblock basis. One such slice 168 is illustratively shown in FIG. 5. Slice 168 is a chain of macroblocks 142. The image 140 may be divided into one or a plurality of slices 168.

Subdivision of an image subdivided into macroblock pair regions, macroblocks, transform blocks, and slices will be described first, and then the functions of the distribution unit 126 and the formation unit 128 will be described in detail. Shall be. As shown in FIG. 5, the scan order defined between the transform coefficients allows the two-dimensionally arranged transform coefficients to be aligned in a linear order with the monotonically increasing frequency content they refer to. do. The distribution unit 126 may convert the transform coefficients of some macroblocks 142 into layers of different quality (ie, zero order layer and refinement layers associated with the data stream 28). To any one of the layers (30). More specifically, the distribution layer 126 increases the SNR quality of the reconstructible video from the respective data streams as the distribution layer number increases from zero or layer 28 to the highest quality refinement layer 30. Attempts to distribute the transform coefficients to the data streams 28, 30. Typically, this will result in low frequency transform coefficients matching low scan positions being distributed to low quality layers, and high quality frequency conversion coefficients being distributed to high quality layers. On the other hand, distribution unit 126 attempts to distribute transform coefficients having a high transform coefficient value to low quality layers, and to distribute transform coefficients having a low transform coefficient value or energy to high quality layers. The distribution performed by distribution unit 126 may be performed in such a way that each transform coefficients are distributed to one layer. However, it is also possible that the distribution performed by the distribution unit 126 is performed by dividing and distributing the sum of the transform coefficients into layers of different image quality so that the sum of the distributed portions reaches the transform coefficient value. Details of other possibilities for dispensing performed by dispensing unit 126 will be described below with reference to FIGS. 6A-6G. Forming unit 128 uses the distribution result from distribution unit 126 to form separate sub-data streams 28, 30. As already mentioned above, the sub-data stream 28 forms the lowest quality refinement sub-data stream and includes the motion data input at input 124. This zero-order sub-data stream 128 may also be divided by the first distributed portion of the transform coefficient values. Thus, the sub-data stream 28 considers an improvement to the higher picture quality of the base-picture layer data stream 26-the higher spatial picture quality-in the embodiment of FIG. 1, while further SNR picture enhancements are sub-. It can be obtained by adding one of the higher quality refinement sub-data streams 30 to the data stream 28. The number of such refinement quality sub-data streams 30 is N, where N may be one or plural. Thus the transform coefficients are " distributed " to these sub-data streams 28, 30 by it-for example, in increasing order of importance for SNR picture quality.

6A shows an example of the distribution of the first 26 transform coefficient values of an 8x8 transform block. More specifically, FIG. 6A shows a table in which the first line of the table lists the individual scan positions according to each scan order (154, 158 in FIG. 5). The illustrated scan positions are typically in the range from 0 to 25. The next three lines show, for the individual transform coefficient values, corresponding distribution values individually mixed with the respective sub-data streams 28, 30. More specifically, for example, the second line corresponds to a zero order sub-data stream 28, while the second line at the end (the third line) is the next high refinement layer (the next). higher refinement layer) 30 and the last line further refers to the next quality layer refinement data-stream. According to the example shown in FIG. 6A, “122” is coded into sub-data streams 128 for the DC component (ie, the transform coefficient value belonging to scan position 0). The distribution values for this transform coefficient with scan position 0 in the next two sub-data streams 30 are set to zero as shown by the hashing of the individual table items. According to the example shown in FIG. 6A, in this manner the zero order enhancement layer sub-data stream 28 is a distribution value for each transform coefficient value. It includes. However, within the transform block of FIG. 6A, only the transform coefficient values of scan positions 0 to 6, 8 and 9 belong to zero order quality layers. The extra transform coefficient values are set to zero. Within other transform blocks, transform coefficient values belonging to the zero order quality layer may belong to different scan positions. Similarly, the transform coefficient values of scan positions 7, 10-12, 15-18 and 21 belong to the next higher quality layer. The residual transform coefficient values are set to zero. The coefficient values of the remaining scan positions are included in the next higher quality layer sub-data stream. As can be seen, certain transform coefficient values can actually be zero. In the example of FIG. 6A, scanning position 23 is such a case. The corresponding distribution values in the above-described image quality layers are set to 0, and the transform coefficient value for the scan position 23 in the last image quality layer (last line) for the scan position 23 is itself zero.

Thus, on the decoder side, for each scan position, various quality layers sub, in order to allow the actual transform block to be reconstructed by summing up distribution values for individual scan positions of different quality layers. The distribution values contained in the bit streams 28, 30 sum up to the actual transform coefficient values.

According to the embodiment of FIG. 6A, each of the sub-data streams 28, 30 individually includes a distribution value for all transform coefficients and all scan positions. But this is not a necessary example. First, as already described above, the zero order sub-data stream 28 does not have to contain any transform coefficients or distribution values. Thus, in the example described below, the last three lines of the table of FIG. 6A are the first refinement layer sub-data streams 30 with the zero order sub-data stream 28 simply containing motion information from the input 124. ) May be seen as included in.

Moreover, the distribution values set to 0 in FIG. 6A and the actual transform coefficient values that are substantially zero are simply used for the use of hashed table entries for the purpose of an easier understanding of the function of the forming unit 128. Distinguished by However, the sub-data streams 28, 30 can be interpreted in such a way that the distinction between the distribution values set to zero just mentioned and the distribution values originally zero is transparent at the decoder. More specifically, some of the respective distribution values for the individual scan positions (ie, numbers from the second to fourth lines below the individual scan position in the first line of FIG. 6A) are set to zero or naturally zero. Reveal transform coefficient values independent of individual distribution values within the sum of.

In the embodiment of FIG. 6A, the forming unit 128 respectively codes one of the sub-data streams 28, 30 with a distribution value for the respective scan positions. This is not essential. For example, according to the embodiment of FIG. 6B, the series of quality layer sub-data streams only contain their transform coefficient values belonging to each quality layer.

The order in which distribution values and transform coefficient values are coded into sub-data streams 28 and 30, respectively, may vary in the embodiments of FIGS. 6A and 6B, respectively. For example, the sub-data streams 28 and 30 may be packetized data streams, where each packet corresponds to one slice 168. Within one slice 168, transform coefficient values may be coded in respective packet macroblock-wise. That is, the scan order is transform coefficient values for the predetermined macroblock 142 fully coded into individual packets before the first transform coefficient value of the macroblock following the macroblock scan order among the macroblocks in the slice 168. Can be defined with Within each macroblock, scan orders may be defined between respective transform blocks 146 and 148 in each macroblock. In other words, before the first transform coefficient value of the next transform block is coded identically, the transform coefficient values are formed in a forming unit (i. 128 may be coded into each of the sub-data streams 28, 30. Within each transform block, coding for transform coefficient values and distribution values may be performed in the manner described with reference to FIG. 8 or 9, respectively, described below.

According to the embodiments of FIGS. 6A and 6B, the transform coefficient values of the other transform blocks of the slice 168 belonging to each one of the image quality layers are spread to other parts of the scan order. More specifically, even though scan positions 0 to 6, 8, and 9 belong to the zero order quality layer within a specific transform block, which is shown by way of example in FIGS. 6A and 6B, a scan belonging to this layer within another transform block. The set of positions can be different. However, according to the embodiment of FIG. 6C, the distribution unit 126 transforms the other transform blocks in the slice 168 in such a way that transform coefficient values of the same set of scan positions for all transform blocks belong to the same quality layer. Distribute count values. For example, in FIG. 6C this is true for all transform blocks in slice 168 and the transform coefficient values in scan positions 0-11 belong to zero order sub-data stream 28.

In addition, according to the embodiment of FIG. 6C, the transform coefficient values belonging to a particular one of the image quality layers are spread over a continuous sequence of consecutive scan positions. But this does not have to be the case. More specifically, transform coefficient values belonging to one scan position between the first and last scan positions belonging to a specific picture quality layer may belong to one of the other picture quality layers as shown in FIG. 6B. However, in the case of the embodiment of Fig. 6c only two syntax elements (one syntax element representing the first scan position of each image quality layer, i.e. scan_idx_start and the last scan position for each image quality layer, are represented. By the use of another syntax element, scan_idx_end, it is possible to represent the scan positions respectively merged with either of the image quality layer sub-data streams 28, 30.

On the one hand the designation of a particular set of scan positions for each one of the picture quality layers and on the other hand the picture-importance dependent on the distribution of the transform coefficients for the respective picture quality layers may be mixed as shown in the following embodiments. have. For example, FIG. 6D shows an embodiment in which the distribution unit 126 distributes transform coefficients to picture quality layers as shown in FIG. 6A. This distribution is different from the transform block. On the other hand, however, each picture quality layer is commonly assigned a particular portion of scan positions for all of the transform blocks. For example, the lowest quality layer is assigned a complete set of scan positions from scan position 0 to scan position 63. Thus, for each transform block, the lowest quality layer contains 64 distribution values. The high quality layer sub-data stream then contains distribution or transform coefficient values for all transform blocks within a particular scan position radius from scan position 6 to scan position 63. The scanning position range of the quality layer then extends from scan positions 13 to 63. In other words, the decoder does not need to know whether a particular one of the distribution values is set to zero (hashed entry) or actually represents a zero transform coefficient value or a meaningless transform coefficient value. However, the decoder needs to know the syntax element scan_idx_start indicating for each slice 168 from any scan position based on the transform coefficient or distribution values that are to be included and used in each sub-data stream. More precisely, for example in the embodiment of FIG. 6D, the sub-data stream corresponding to the second to last line includes transform coefficients or distribution values for the individual transform block 58. In the case of the transform block of FIG. 6D, the second value is 22 while the first value is 0. FIG. The first transform coefficient value of each image quality layer corresponding to scan position 6 is known by using the syntax element scan_idx_start at the decoder, while the residual transform coefficient values of the image quality layer refer to the following scan positions. Similar to the embodiment of FIG. 6D, FIG. 6E is a syntax element scan_idx_end indicating the last scan position up to each quality layer sub-data stream including sub-coefficients or distribution values for individual sub-data streams. An example is shown.

The combination of the embodiments of FIGS. 6D and 6E is shown in FIG. 6F. According to this embodiment, each set of scan positions belonging to a particular one of the picture quality layers extends from the first scan position indicated by the syntax element scan_idx_start to the last scan position indicated by the syntax element last_idx_end. For example, within the image quality layer corresponding to the second line from the end, each set of scan positions extends from scan position 6 to scan position 21. Finally, the embodiment of FIG. 6G focuses on the embodiment of FIG. 6C according to the distribution of the individual transform coefficient values of the other transform blocks in slice 168 where the use of syntax elements scan_idx_start and / or scan_idx_end is common for the transform blocks. It can be combined with. Thus, according to the embodiment of FIG. 6G, within a particular one of the image quality layers, all the transform coefficient values from scan_idx_start to scan_idx_end are distributed to each image quality layer. Thus, unlike the embodiment of FIG. 6F, in the embodiment of FIG. 6G, all transfer coefficient values from scan position 6 to scan position 21 are assigned to the image quality layer corresponding to the second line at the end of FIG. 6G. do. Alternatively, in the embodiment of FIG. 6F, some of the distribution values within this 6 to 21 position scan range may be set to 0, where a transform coefficient set to 0 within this 6 to 21 position scan range. The distribution of values and transform coefficient values not set to zero may be different than any one of the other transform blocks in the current slice.

Next, the cooperation between the hybrid coder 42, the layer coding unit 44, the distribution unit 126 and the forming unit 128 is exemplarily described in FIG. 7, with sub-data streams 28 and 30 respectively. An example of the structure is shown. According to the embodiment of Fig. 7, the forming unit 28 is designed such that the individual sub-data streams 28, 30 are each packetized, that is, made up of one or more packets. Specifically, the forming unit 128 is each designed to generate a packet for each slice 168 in the image 140 in each sub-bit stream 28, 30. As shown in FIG. 7, a packet consists of one slicehead 170 and the remaining data 172 on the other, except for a sub-bit stream 28 that selectively constitutes a slicehead within one of each packet. .

A description of the residual data 172, i.e., the residual data # 1, # 2, ..., #N, is made with reference to the above description of Figures 6A-6G, for example the second to fourth lines of this table. Will correspond to the residual data # 1, # 2, # 3. In other words, the residual data 127 shown in FIG. 7 also includes the transform coefficient values discussed in FIGS. 6A-6G, wherein the distribution of the transform coefficient values in each of the sub-data streams 28, 30 is not described herein again. Do not. However, FIG. 7 further shows the syntax element included in slicehead 170 and residual data 172 from hybrid code 42. As described above, the hybrid coder 42 is opened or closed on the basis of macroblocks between several inter-layer prediction modes to expect motion information from the base layer, or coded to expect motion information from the base layer as motion information as residuals. Or generate new motion information for each motion block of the high refinement layer that newly codes the motion information. Therefore, as shown in FIG. 7, the residual data 172 may be used for the syntax element representing a motion parameter for each macroblock, a macroblock mode such as a field or framecode, or reuse of motion parameters of the base layer of each macroblock. Inference mode can be made. This is especially true for zero or sub-data stream 28. However, since the motion information is not refined again in the next refinement layer and the next high-quality sub-data streams 30 _{1 and} 30 _N , the forming unit 128 may execute macroblock-type syntax elements, motion parameters, sub-data stream (30 _1, ~ 30 _N) of design to leave the inference mode indication in the residual data or the sub-syntax elements which one, of the data streams (30 _1, ~ 30 _N) is a macro-block mode and the sub- Design the same settings as the motion parameters for each macroblock included in data stream 28, or indicate an inference mode for each macroblock to indicate that the same setting is used in each refinement layer . According to an embodiment of the present invention, various sub-data streams (28, 30 ₁ to 30 _N), all the residual data 172 within is processed by using the same syntax structure refinement treatment sub-data stream (30 ₁ to 30 Residual data in _N ) also consists of information defined in a macroblock fashion in a macroblock, motion parameters and / or inference mode activation / deactivation.

Also, as can be derived from FIG. 7, the forming unit 128 may be designed to supply the syntax element scan index start and scan index end to the slice header 170. Alternatively, the slice header 170 may be comprised of different syntax elements that define each individual slice or packet, which is a set of scan locations associated with the residual data corresponding to each slice header data. As already shown above, if the sub-data stream 28 is not composed of any residual data but is only composed of macroblock mode and / or motion parameter and inference mode instruction, respectively, the slice header of the packet of the sub-data stream 28 The data may not consist of syntax elements related to defining layer specific scan locations. Moreover, as already mentioned above, the slicehead 170 may only consist of one of scan_idx_start and scan_idx_end. As a result, scan_idx_start and / or scan_idx_end are provided once in the transform block size category, i.e. every 4x4 and 8x8, or in general for all block size categories for measurements made respectively to convey scan_idx_start and scan_idx_end at different block sizes, described below. It can only be provided once for each slice / image / sub-data stream for that.

In addition, the slice header data may be composed of syntax elements representing the image quality level. For this reason, the forming unit 128 is designed so that a syntax element or an image quality indicator distinguishes only one zero-order image quality level 28 from the other refinement layer 30 ₁ to 30 _N. Optionally, the picture quality indicator may distinguish all picture quality layers from the refinement layers 28 and 30 ₁ to 30 _N. In the latter two cases, the picture quality indicator may omit the macroblock mode, the motion parameter and / or the speculation mode in the packet of the sub-data streams 30 ₁ to 30 _N defined by some macroblock method, because in this case the decoder side This is because such refinement layer sub-data streams 30 ₁ to 30 _N are only known to refine transform coefficients using macroblock mode, motion parameters, and inference mode from zero mode sub-data stream 28. .

Although not described in more detail above, the forming unit 28 is intended to entropy code packets within the sub-data streams 28 and 30 ₁ to 30 _N. 8 and 9 show a feasible example for coding transform coefficients in residual data belonging to one transform block according to the two embodiments. FIG. 8 shows the pseudo code of the first embodiment for feasible coding of transform coefficients in a transform block in some residual data 172. Assuming, the following examples apply:

Scan position 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Counting number 0 One 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Transform coefficient level 7 6 -2 0 -One 0 0 One 0 0 0 0 0 0 0 0

Based on this example, in the following, the pseudo code of FIG. 8 is described showing how the forming unit 128 codes the remaining data in one transform block into certain sub-data streams.

In order to convey the information of the transform coefficients, according to FIG. 8, a parameter coeff_token 240 is first provided. The parameter coeff_token is the number of nonzero coefficients, i.e. total_coeff (coeff_token) and the number of transform coefficients in the execution of transform coefficients with an absolute value equal to one at the end of the order of nonzero transform coefficients, i.e. trailing_ones (coeff_token). Is a code word indicating. In this example, total_coeff (coeff_token) is 5 (transformation coefficient numbers 0, 1, 2, 4 and 7) and trailing_ones (coeff_token) is 2 (transformation coefficient numbers 4 and 7). Therefore, by providing the parameter coeff_token 240, the positions of significant transform coefficients are set to only the total_coeff (coeff_token) range in which non-zero transform coefficients are present.

Then, the values of the levels of these nonzero transform coefficients are provided. This is done in reverse scan order. More specifically, it is first checked as to whether the total number of nonzero transform coefficients is greater than zero (242). This is true in the example above because total_coeff (coeff_token) is 5.

The nonzero transform coefficients are then performed in steps through an inverse scan order (244). The reverse scan order is not yet apparent from merely reviewing the counting parameter increment i ++ in the for-loop 244 but will be apparent from the following evaluation. While stepping through these non-zero transform coefficients in inverse scan order, for the first of these non-zero transform coefficients, only their transform coefficient signal is provided 248. This is done for the first number of trailing_ones (coeff_token) of non-zero transform coefficients when stepping them out in reverse scan order, and for these transform coefficients the absolute value of these transform coefficients is described by trailing_ones (coeff_token) Because it is already known. The coefficient signals thus provided are used for temporary storage in the auxiliary vector coefficient level [i] for the transform coefficient levels of the non-zero transform coefficient levels with an absolute value of 1, where i is nonzero when scanned in reverse scan order. Numbering of the transform coefficients (250). In this example, after the first two cycles of the for-loop 244, level [0] = 1 and level [1] =-1 are obtained.

Next, the coefficient levels coeff_level for the remaining nonzero transform coefficients are provided in reverse scan order (252) and temporarily stored in auxiliary vector coefficients level [i] (254). The remaining for-loop cycles result in level [2] =-2, level [3] = 6 and level [4] = 7.

Now, to uniquely determine the positions of significant transform coefficients, if total_coeff (coeff_token) does not already match the maximum number of transform coefficients in the transform block, i.e. not equal to maxNumCoeff, two additional parameters called total_zeros and run_before are provided. do. More specifically, it is checked as to whether total_coeff (coeff_token) is equal to maxNumCoeff (256). If this is not the case, the parameter total_zeros is provided (258) and the auxiliary parameter zerosLeft is initialized to the value of total_zeros (260). The parameter total_zeros is specified as the number of zeros between the last nonzero count in the scan order and the start of the scan. In the above example, total_zeros is 3 (coefficients number 3, 5 and 6). Therefore, zerosLeft is initialized to 3.

For each of the non-zero transform coefficients except the last one (coefficient number 0) for the inverse scan order, starting at the last non-zero transform coefficient (coefficient number 7) for the scan order (62), as seen in the scan order A parameter run_before is provided that indicates the length of the execution of zero-level transform coefficients arranged directly before each non-zero transform coefficient (64). For example, for i corresponding to 0, the last nonzero transform coefficient for the scan order is the nonzero transform coefficient in the query. In this example, this is a transform coefficient with number 7 and level 1. The execution of zeros before this transform coefficient has a length of two, that is, transform coefficients 5 and 6. Therefore, in this example, the first run_before-parameter is 2. This parameter is temporarily stored in the auxiliary vector coefficient run [0] (266). This is repeated in reverse scan order for run [i] with the count of nonzero transform coefficients when scanned in reverse scan order. In each cycle of the for-loop 261, it is determined for each cycle as to how many zero-level transform coefficients are left by reducing the auxiliary parameter zerosLeft by the parameter run_before. If zerosLeft is zero, no run_before-parameter is no longer provided (270) and the residual coefficients of the vector run are fixed to zero (272). In any case, no run_before-parameter is provided for the last non-zero transform coefficient when stepped in reverse scan order, ie no run_before-parameter for the first non-zero transform coefficient for the scan order is provided. This parameter is derived from the number of zero-level transform coefficients, 274, as represented by the auxiliary parameter zerosLeft. In this example, the auxiliary vector coefficients of the vector run are run [0] = 2, run [1] = 1, run [2] = 0, run [3] = 0 and run [4] = 0.

Finally, in the for-loop pointed out at 276, the values of the transform coefficient levels when stored in the auxiliary vector level are stored in their positions by copying the values of the vector level coefficients for each position in the one-dimensional array coeffLevel. Is assigned. More specifically, in the first iteration of for-loop 276, coeffNum initialized from i = 4 to 0 (278) is increased by run [4] + 1 = 0 + 1 = 1 resulting in coeffNum = 0. CoeffLevel [0] is assigned to the value level [4] = 7. This is repeated for the next auxiliary vector coefficients level [3] through level [0]. All transform coefficients are coded because the remaining positions of the array coeffLevel are initialized to a value of 280 (280).

The syntax elements written in bold in FIG. 8 will be coded into each sub-data stream, for example by variable length coding.

9 is another example of coding of a transform block. In this example, the scanning order clarifies itself in " i ++ " in the while-loop 310 indicating that the counting parameter i is incremented per while-loop iteration.

For each coefficient in the scanning order, a 1-bit symbol significant_coeff_flag is provided (312). If the significant_coeff_flag symbol is 1 (314), ie if a nonzero coefficient is present at this scanning position i, then the 1-bit symbol last_significant_coeff_flag is provided (316). This symbol indicates whether the current significant coefficient is the last one in the block or, moreover, the important coefficients follow the scan order. Therefore, if the last_significant_coeff_flag symbol is 1 (318), this indicates that the number of coefficients, ie numCoeff is i = 1 (320), is inferred (322) so that the level of the next transform coefficients is zero. Up to now, the syntax elements last_significant_coeff_flag and significant_coeff_flag can be represented as important maps. Then, for the last transform coefficient in the scanning order, an absolute value of level minus 1, i.e. coeff_abs_level_minus1 and its signal, coeff_sign_flag, is provided (324), thus representing the transform coefficient level of this last significant transform coefficient (326). ). These steps 324 and 326 are repeated for the residual significant 330 transform coefficients in the inverse scanning order (332) (328), where the inverse scan order is itself apparent at i--, i. And the counting parameter i is decreased from cycle to cycle in the for-loop. The parsing of the syntax elements coeff_abs_level_minus1 starts with the binarization derivation for the possible values of the syntax element. The binarization scheme may be UEG0, that is, a concatenated unary / zero-th order Exp-Golomb binarization process. Depending on the possible binarization, each syntax element can be binary arithmetically coded as bin by bin. In this regard, the context adaptive binary arithmetic coding scheme can be used for the prefix part of the binarization of coeff_abs_level_minus1 while using a decimal bypass without adaptation for any suffix part.

For completeness, it is noted from FIG. 5 that the number of distinguishable scan positions in the 8 × 8 transform block is 64 while the number of distinguishable scan positions in the 4 × 4 transform block is only 16. Thus, the syntax elements scan_idx_start and scan_idx_end mentioned above can be defined with an accuracy that can be distinguished between all 64 scan positions or between only 16 scan positions. In the latter case, for example, syntax elements may be applied to each four times successive transform coefficients in an 8x8 transform block. More precisely, 8x8 transform blocks can be coded by using:

residual_block (LumaLevel8x8,

4 * scan_idx_start, 4 * scan_ids_end +3, 64),

And in the case of 4x4 transform blocks it can be coded by the following use:

residual_block (LumaLevel4x4,

scan_idx_start, scan_idx_end, 16).

Here, residual_block becomes residual_block_cavlc or residual_block_cabac, and LumaLevel4x4 and LumaLevel8x8 represent an array of luma samples of 4x4 and 8x8 transform blocks, respectively. As can be seen, scan_idx_start and scan_idx_end are specified such that they are distinguished between 16 scan positions in order to accurately represent the range of positions in 4x4 blocks. By the way, in an 8x8 block, the accuracy of these syntax elements is not sufficient to adjust the range in these blocks in a four-fold manner.

In addition, 8x8 blocks of transform coefficients also divide the 64 coefficients of the 8x8 block into four sets of 16 coefficients, for example, starting at coefficient n with n in the range including 0-3, for example every fourth coefficient. by n-th set, and by encoding each set of 16 coefficients using the residual block configuration for 4x4 blocks. On the decoder side, four sets of these 16 coefficients are recombined to form one set of 64 coefficients representing an 8x8 block.

After describing an embodiment for the encoder, a decode for decoding each of the image quality scalable data streams is described in FIGS. 10 and 11. 10 shows the general structure of the decode 400. Decode 400 includes a demultiplexer 402 having an input 404 for receiving a scalable bit stream 36. Demultiplexer 402 demultiplexes input signal 36 into data streams 26-32. Because of this, the demultiplexer performs decoding and parsing functions, or decodes and parses functions. For example, demultiplexer 402 decodes the transform block coding of FIGS. 8 and 9. Recall also Figures 6A-6G. Thus, the demultiplexer 402 uses the sub-data stream information to parse the current sub-data stream to know how transform coefficient values or distribution values for a particular transform block are expected. The retrieved data stream is thus obtained by decoding unit 406, which reconstructs video 18 based on this data stream and at each output 410 each reconstructed video 408. )

The internal structure of the decoding unit 406 is shown in more detail in FIG. As shown, the decoding unit 406 includes a base layer motion data unit 412, a base layer residual data input 414, a zero-order refinement layer motion data input 416, and an optional transform coefficient zero-order refine. A transform transform coefficient data input 418 and an input 420 for the sub-data stream 30. As shown, inputs 412 and 414 are for receiving data stream 26, while inputs 416 and 418 are for receiving data stream 28. In addition, the decoding unit 406 includes a low quality reconstructed video signal output 422, a high quality interlayer coded reconstructed video signal output 424, and an internally coded reconstructed video signal output 426, The latter provide information about high definition video signals.

The combiner 428 has an output connected to the inputs 418 and 420 and an output for outputting the transform coefficient levels for each of the transform blocks obtained by collecting distribution values corresponding to the various quality layers. The collection involves the sum of the distribution values for a particular transform coefficient in some data streams 30-28. However, it is also possible for combiner 428 to adjust all the transform coefficient values to zero in advance and only return to zero if the distribution value for each scan position among these zeros is not equal to zero. In this way, the combiner collects information about the transform coefficients of the various transform blocks. The combination of expected or transform coefficient values in each layer is associated with the combiner using the scan position information of the current layer, such as scan_idx_start or scan_idx_end. Optionally, the combiner uses the information of the transform coefficient values in the respective transform blocks received so far from the low quality to the SNR layer.

The transform block output by combiner 428 is received by residual prediction decoder 430 and adder 432.

Between the residual prediction decoder 430 and the input 414, a later or inverse transform unit 432 is connected to send the residual data inversely transformed to the residual prediction decoder 430. The latter uses the inverse transformed residual data to obtain a predictor that is eventually added to the transform coefficients of the transform block output by combiner 428 after performing up-sampling or other quality adaptation. On the other hand, motion prediction unit 434 is coupled between input 412 and input of adder 436. The input of another adder 436 is connected to the output of inverse transform unit 432. In this way, motion prediction unit 434 uses motion data at input 412 to generate a prediction signal for the residual signal output inversely transformed by inverse transform unit 432. The result of adder 436 at the output of adder 436 is the reconstructed base layer video signal. The output of adder 436 is not only connected to the input of prediction decoder 432 but also to output 432. Prediction decoder 432 consequently uses the reconstructed base layer signal to expect the intra-layer coded portion of the video content output by combiner 428 using up-sampling. On the other hand, the adder 436 output causes the motion prediction unit 434 to generate a prediction signal on the second input of the adder 436 based on the signal reconstructed from the base layer data stream using the motion data at the input 412. Is also connected to the input of motion prediction unit 434. The expected decoded transform coefficient value output by residual prediction decoder 430 is inverse transformed by inverse transform unit 438. At the output of the inverse transform unit 438, high quality residual video signal data is generated. This high quality residual data video signal is added by the adder 440 with the motion predicted video signal output by the motion prediction unit 442. At the output of the adder 440 a reconstructed high quality video signal appears which is coupled to the output 424 and also to the input of the motion prediction unit 442. Motion prediction unit 442 is not only reconstructed video signal output by adder 440 but also motion information by motion parameter prediction decoder 444 coupled between input 416 and the input of each motion prediction unit 442. Perform motion prediction based on the output. The macroblock selective based motion parameter prediction decoder 444 uses the motion data from the base layer motion data input 412 as a predictor and outputs the motion data to the motion prediction unit 442 depending on this data, for example. The motion vector at input 416 is used as an offset vector to the motion vector at input 412.

The embodiment described above enables an increase in the granularity of SNR scalable coding at the image / slice level compared to the CGS / MGS coding described in the introduction, but with a significant increase in the complexity present in FGS coding. There is no. In addition, bitstream adaptation is possible by simple packet dropping, due to the fact that truncated packets in FGS are not widely used.

The above-described embodiment has the basic idea to split the traditional CGS / MGS packet transform coefficient level, which is usually specified in the SVC draft, generally into subsets that are delivered to other packets and to other SNR refinement layers. As an example the embodiment described above relates to a CGS / MGS coding one base and one enhancement layer. These data are different in each image to increase the granularity of SNR scalable coding instead of the enhancement layer containing all macroblocks as well as macroblock mode, internal prediction mode, motion vectors, reference picture indices, and other control parameters. It will be distributed to slices, other packets, and other enhancement layers. In the first enhancement layer, the macroblock mode, motion parameters, other control parameters as well as optionally the first subset of transform coefficient levels are delivered. In the subsequent enhancement layer, the same macroblock mode and motion vector are used, but the second subset of transform coefficient levels is encoded. All transform coefficients already transformed in the first enhancement layer may be adjusted to 0 in the second and all subsequent enhancement layers. In the third and subsequent enhancement layers, the macroblock mode and the motion parameters of the first enhancement layer are used again, but subsequent subsets of the transform coefficient levels are encoded.

It should be noted that this division does not increase or increase only slightly in complexity compared to the traditional CGS / MGS coding specified in the current SVC draft. All SNR enhancements can be analyzed in parallel and the transform coefficients need not be collected from other scans for the image / slice. That means, for example, that a decoder could analyze all transform coefficients for a block from all SNR enhancements and apply an inverse transform on the block without storing the transform coefficient levels in a temporary buffer. When all blocks of the macroblock have been fully analyzed, the motion compensated expectations are applied and the last reconstruction signal for this macroblock can be obtained. It should be noted that all syntax elements of the slice are transmitted by the macroblock to the macroblock and within the macroblock the transform coefficient values are sent by the transform block to the transform block.

Flags can be encoded at the slice level, whether or not in all macroblock modes, this signal and motion parameters are inferred from the base layer. The current syntax of a given CGS / MGS packet is equal to 0 without all the syntax elements mb_skip_run and mb_skip_flag transmitted, and all syntax elements mb_field_decoding_flag equal to their values in the base layer macroblocks that are co-located without being transmitted. This means that the syntax elements base_mode_flag and residual_prediction_flag are not transmitted but are inferred to be equal to one. In the first SNR enhancement layer, this flag is usually set to 0 because it is possible to transmit a motion vector different from the base layer in order to improve the coding efficiency. However, in all other enhancement layers, the flag is set to 1 because these enhancement layers only represent refinement of the transform coefficient level at the scan position that was not coded in the previous SNR enhancement layer. Coding efficiency can be improved by setting this flag to 1 because unnecessary syntax elements are not sent and therefore the associated bitrate is saved.

Furthermore, as described above, among the various transform blocks, the first scanning position x for the transform coefficient level may be transmitted at the slice level and no syntax element may be passed at the macroblock level for the transform coefficient for the scanning position smaller than x. have. In addition to the above described that the first scanning position is transmitted only for a specific conversion size and the other scanning size is inferred based on the converted value, it is possible to transmit the first scanning position for all supported conversion sizes. Do.

Similarly, the last scanning position y for the transform coefficient levels in the various transform blocks may be sent at the slice level rather than at the macroblock level for the transform coefficients for the scanning positions greater than y. In addition, it is possible to send the last scanning position for all supported conversion sizes or only to send the last scanning position for a specific conversion size, and to deduce the last scanning position for another conversion size based on the converted value. .

The first scanning position for each transform block in the SNR enhancement layer can be selectively inferred based on the transform coefficients sent to the previous enhancement layer. This law of inference is applied to all transform blocks independently and the other first transform coefficients in each block are derived, for example, by combiner 428.

Further combinations are made to signal and infer the first scanning position. This means that the first scanning position is basically inferred based on the level of transform coefficients already transmitted in the previous SNR enhancement layer, but for this purpose the additional scanning position cannot be smaller than the value x transmitted in the slicehead. Is used. With this concept it is again possible to have a different first scanning index for each transform black, which is chosen to maximize coding efficiency.

As a further alternative, signaling of the first scan position, inference of the first scan position, or a combination thereof is combined with signaling the last scanning position.

What is described above allows a possible scheme that allows for SNR scalability, where only subsets of transform coefficient levels are transmitted in different SNR enhancement layers, in which one mode Examples are signaled by more slicehead delimiters, macroblock mode and motion parameters deduced for all macroblock types, and / or no transform coefficients for some scanning positions at the transform block level. . The slice level syntax element may use signals inferred from base layer macroblocks in which the macroblock mode and motion parameter for all macroblocks lie together. Specifically the same macroblock mode and motion parameters may be used and the corresponding syntax elements may not be transmitted at the slice level. The first scanning position x for every transform block can be signaled by the slicehead syntax element. A syntax element is not transmitted for transform coefficient values of scanning positions less than x at the macroblock level. Optionally, the first scanning position for the transform block is inferred based on the transmitted transform coefficient level of the base layer. Combinations of both of the latter are also possible. Similarly, the last scanning position y for all transform blocks can be signaled by the slicehead syntax element and no syntax element is passed for the transform coefficient value of the scanning position greater than y at the macroblock level.

As noted above, the details described in FIGS. 1-11 can be varied in many ways. For example, although the above embodiment can be an example of a two spatial layer environment, the above embodiment can easily be transformed into one or more quality layers with only one quality layer or only N + 1 SNR scalable refinement layers. I can deliver it. For example, assume that 12 parts are missing from FIG. In this case the hybrid coder 42 causes the video sub-data stream 28 to include scan area information in which the unit 44 represents each of various image quality layers, video sub-data streams 30, or a subset of possible scan positions. +30), obtaining transform blocks 146 and 148 of transform coefficient values for image 140 of the video signal while acting as a means of forming transform coefficient information for transform coefficient values belonging to a subset of possible scan positions. Act as a coding means for coding the video signal 18 using block transform. Inner layer prediction is not relevant. In addition, coder 42 is simplified to only perform blockwise transformation without motion prediction. Similarly, in the case of a quality layer, the demultiplexer 402 acts as a parsing means for parsing video sub-data streams of a plurality of quality layers to obtain scan area information and transform coefficient information for each quality layer. An inverse transform unit 438 reconstructs an image of a video signal by inverse transforming the transform block from the transform coefficient information to a subset of possible scan positions while using the scan area information. By associating with, it acts as a means of creating a transform block for each quality layer.

In addition, the embodiment of FIG. 1 can be changed in such a way that the base layer coder 12 acts at the same spatial resolution as the bit depth as the enhancement layer code 14. In that case the embodiment shows the SNR scalable coding as the standard base layer 26 and the various enhancement layers 28 and 30, including the division of the transform coefficients.

Depending on the actual implementation, the invention summary may be implemented in a hardware or software manner. Therefore, the present invention may also be stored on a CD or disk or other data carrier which is a computer readable means associated with the computer program. Therefore, the present invention is also a computer program having a program code, so that the present invention can be carried out in connection with the above form when executed in a computer.

In addition, all the steps and functions shown in the flowchart can be implemented by the respective means in the encoder, and the implementation may consist of subroutines executed in the CPU.

FIELD OF THE INVENTION The present invention relates to picture quality-scalable video data streams and is applied to video data stream generation and decoding.

Claims (41)

In each transform block, for each possible scan position, at least one of the transform coefficient values in each transform block has the preset scan positions 154, 156, 164, 166 with the possible scan positions such that they belong to each possible scan position. Is determined between transform coefficient values in the transform block, and the video signal using block-wise transform to obtain transform blocks 146 and 148 of transform coefficient values for the image 140 of the video signal. Coding means 42 for encoding 18); And Transform coefficient information for transform coefficient values belonging to the sub-set of the possible scan positions such that the sub-set of each quality layer includes at least one possible scan position not covered by the other multiple sub-sets of quality layers. And forming means (44) for forming, for each of the plurality of image quality layers, a video sub-data stream (30; 28, 30) comprising scan area information indicative of the sub-set of possible scan positions; Apparatus for generating a quality-scalable video data stream comprising a. The method according to claim 1, The forming means 44 is configured such that the scan area information is configured to indicate a first scan position among the possible scan positions in the sub-set of the possible scan positions in the preset scan order. Flexible video data stream generating device. The method according to claim 1 or 2, The forming means 44 is formed such that the scan area information is configured to indicate a last scan position among the possible scan positions in the sub-set of the possible scan positions in the preset scan order. -Apparatus for generating scalable video data stream. The method according to any one of claims 1 to 3, The coding means 42 includes at least one first transform block 148 having a 4 × 4 transform coefficient value and at least one second transform block 146 having an 8 × 8 transform coefficient value among the transform blocks. And wherein according to the preset scan order, each of the 4 x 4 transform coefficient values belongs to the other of the possible scan positions, while among the 8 x 8 transform coefficient values, four consecutive And the transform coefficient values belong to the same possible scan position. The method according to any one of claims 1 to 4, The forming means 44 converts the transform coefficient values belonging to the sub-set of the possible scan positions into the transform coefficient information in a block manner such that the transform coefficient values of a predetermined transform block form a continuous portion of the transform coefficient information. Apparatus for generating a quality-scalable video data stream, characterized by being encoded. The method according to claim 5, The forming means By encoding into the video sub-data stream a significant map (significant_coeff_flag, last_significant_coeff_flag) belonging to the sub-set of the possible scan positions and specifying positions of the non-zero transform coefficient values in a preset transform block, and then The possible in the preset transform block in an inverse scan order opposite to the preset scan order, which belongs to a subset of the possible scan positions in the preset transform block and starts at the last non-zero transform coefficient value. By encoding the transform coefficient values that belong to a sub-set of scan positions and are nonzero Picture quality scalable video data streams, characterized in that for encoding the continuous part. The method according to claim 6, The forming means Significant flags (significant_coeff_flag) depend on the respective transform coefficient values that are zero or non-zero, and in the preset order, nonzero and the possible from the first transform coefficient value belonging to the sub-set of possible scan positions. By encoding the significant flag (significant_coeff_flag) into the video coding stream for transform coefficient values belonging to the sub-set of possible scan positions up to the last transform coefficient value belonging to a sub-set of scan positions, and After each significant flag of each non-zero transform coefficient value, depend on the respective transform coefficient value that is the last transform coefficient value belonging to the sub-set of the possible scan positions in the preset transform block that is non-zero or zero. By encoding the last-flag (last_significant_coeff_flag) And an image quality-scalable video data stream generation device configured to encode the critical map. The method according to any one of claims 1 to 7, The preset scan order is to scan the transform coefficient values of the transform blocks such that transform coefficient values belonging to a higher scan position in the preset scan order correspond to higher spatial frequencies. Data stream generator. The method according to any one of claims 1 to 5, The forming means The preset transform as well as the number of consecutive trailing transform coefficient values belonging to the sub-set of the possible scan positions within the preset transform block and having an absolute value of one of the nonzero transform coefficient values By encoding significant information belonging to the sub-set of the possible scan positions in the block and specifying a number of nonzero transform coefficient values; By encoding the successive trailing transform coefficient values and signals and signals of non-zero residual transform coefficient values belonging to the sub-set of the possible scan positions in the preset transform block; All of the transform coefficient values that are 0 out of the first transform coefficient values belonging to the sub-set of possible scan positions to the last transform coefficient values that belong to the sub-set of the possible scan positions in the preset transform block and that are nonzero By encoding; By encoding a number of consecutive transform coefficient values belonging to the sub-set of the possible scan positions in the preset transform block and immediately preceding any of the number of nonzero transform coefficient values in the inverse scan order, And an image quality-scalable video data stream generating device configured to encode the continuous part. The method according to any one of claims 1 to 9, The coding means applies the blockwise transformation using motion-prediction and followed by quantization of the motion-prediction residual and the transform block for the motion-prediction residual on the image of the video signal to obtain motion information. And to encode the video signal. The method according to claim 10, The forming means 44 is To distribute the transform coefficient values to different image quality layers such that different values of the transform coefficient values are distributed to different image quality layers within a separate block of the transform blocks; And The indication in the sub-data stream of the first quality layer includes the motion information such that the sub-data stream of the first layer of the quality layers contains the motion information and has an indication indicating the presence of motion information. Indicate the motion information non-existence as part of the quality-scalable video data stream in addition to the sub-data streams, and Such that the sub-data streams of the other picture quality layer (s) have an indication indicating the absence of motion information, or have an indication indicating the presence of motion information and contain inadvertent motion information. So that each sub-data stream includes transform coefficient information for the transform coefficient values distributed to each of the quality layers, as well as an indication of the presence or absence of motion information for each quality layer. For a quality layer, to form one of the sub-data streams; Apparatus for generating a quality-scalable video data stream, characterized in that formed. The method according to claim 11, The coding means is configured to hybrid-encode a low quality version of the video signal to obtain low quality motion information, and to hybrid-encode a high quality version of the video signal to obtain high quality motion information, and The forming means is characterized in that the motion information is equal to the high quality motion information or refinement information for allowing reconstruction of the high quality motion information based on the low quality motion information. An apparatus for generating a quality-scalable video data stream, wherein a sub-data stream has an indication indicating the presence of the motion information, and wherein a portion of the quality-scalable video data stream is also configured to include the low quality motion information. . The method according to claim 11 or 12, And said coding means and said forming means are formed such that said motion information and said display are related to a macroblock of said video. The method according to any one of claims 1 to 13, The forming means Within individual blocks of the transform blocks, to distribute the transform coefficient values to other image quality layers such that different values of the transform coefficient values are distributed to other image quality layers, and Said one of said sub-data streams being distributed to said respective quality layers so that each sub-data stream can be parsed independently from said other sub-data stream (s) independently to obtain said respective transform coefficient information. To form one of the sub-data streams, for each quality layer, to include transform coefficient information for transform coefficient values Apparatus for generating a quality-scalable video data stream, characterized in that formed. The method according to claim 14, The forming means for at least one sub-data stream, wherein the transform coefficient information of the at least one sub-data stream comprises the transform coefficient values belonging to each sub-set of the possible scan positions, and the at least And the transform coefficient values belonging to the sub-set of the possible scan positions in addition to the transform coefficient values distributed in the image quality layer of one sub-data stream, are configured to be fixed to zero. . The method according to claim 14 or 15, Said forming means being configured to combine said transform coefficient values and said respective transform coefficient information distributed to respective picture quality layers independent of said other sub-data stream (s). -Apparatus for generating scalable video data stream. The method according to claim 16, The forming means is characterized in that the sum of the distribution values is equal to one transform coefficient value and the transform coefficient information of one or more layers of the image quality layers is configured to include one distribution value with respect to the one transform coefficient value. Image quality-scalable video data stream generating device. The method according to claim 17, Wherein said forming means has said distribution value associated with said one transform coefficient value contained by said transform coefficient information of said image quality layer to which said one transform coefficient value is distributed equal to said one transform coefficient value and another distribution value. (S) is configured to be zero to generate a quality-scalable video data stream. The method according to claim 14 or 15, The forming means is independent of the sub-data stream (s) of the quality layer (s) followed by the sub-data stream of the first quality layer in the layer order such that a layer order is defined between the quality layers. In which the sub-data stream (s) of the subsequent picture quality layers in the layer order are combined with the transform coefficient values distributed to the first picture quality layer in the layer order. And are configured to be combined only with the sub-data stream (s) of the image quality layer (s) preceding each image quality layer so as to be combined with the respective transform coefficient information and the transform coefficient values distributed to the respective image quality layers. An image quality-scalable video data stream generating device. The method according to any one of claims 1 to 19, And said forming means is arranged such that said sub-data streams are arranged in packets, with one packet per slice. 10. An apparatus for reconstructing a video signal from a picture quality-scalable video data stream 36 comprising a video sub-data stream 30 (28, 30) for each of a plurality of picture quality layers. In each transform block, for each possible scan position, at least one of the transform coefficient values in each transform block has a preset scan order with the possible scan positions so that it belongs to each possible scan position. Determined for each image quality layer, to obtain, for each image quality layer, transform coefficient information and scan region information for transform coefficient values of different transform blocks, and the scan region information indicating a subset of the possible scan positions, wherein Parsing means 402 for parsing video sub-data streams of a plurality of quality layers; Constructing the transform block using the scan area information by combining the transform coefficient values of the respective transform blocks from the transform coefficient information with the sub-set of the possible scan positions, for the respective picture quality layers. Means 428; And Reconstruction means (438) for reconstructing the image of the video signal by inverse transformation of the transform blocks; And a video signal reconstruction device from a picture quality-scalable video data stream. The method according to claim 21, The building means 428 uses the scan area information as if the scan area information indicates a first scan position among the possible scan positions in the sub-set of the possible scan positions in the preset scan order. And video signal reconstruction from a picture quality-scalable video data stream. The method according to claim 21 or 22, The building means 428 to use the scan area information as if the scan area information indicates a last scan position among the possible scan positions within the sub-set of the possible scan positions in the preset scan order. And a video signal reconstruction device from a picture quality-scalable video data stream. The method according to any one of claims 21 to 23, When the parsing means 402 and the building means 428 are between the transform blocks, the at least one first transform block 148 consists of 4 × 4 transform coefficient values and at least one second The transform block 146 consists of 8 x 8 transform coefficient values, and then, according to the preset scan order, each of the 4 x 4 transform coefficient values belongs to the other of the possible scan positions, while the 8 Between x 8 transform coefficient sets, four consecutive transform coefficient values are configured to belong to the same possible scan position. The method according to any one of claims 21 to 24, The parsing means 402 is configured such that the transform coefficient values belonging to the sub-set of the possible scan positions can be encoded in the transform coefficient information in a block manner. Rebuild device. The method according to claim 25, The parsing means 402 is By decoding into the video sub-data stream a significant map (significant_coeff_flag, last_significant_coeff_flag) belonging to the sub-set of the possible scan positions and specifying positions of the transform coefficient values other than zero in the preset transform block, and then The possible in the preset transform block in an inverse scan order opposite to the preset scan order, which belongs to a subset of the possible scan positions in the preset transform block and starts at the last non-zero transform coefficient value. By decoding the transform coefficient values belonging to a sub-set of scan positions and being nonzero And a video signal reconstruction from a picture quality-scalable video data stream. The method of claim 26, The parsing means Significant flags (significant_coeff_flags) depend on respective transform coefficient values of zero or nonzero, and in the preset order, nonzero and the possible scan from the first transform coefficient value belonging to the sub-set of possible scan positions. By decoding the significant flag (significant_coeff_flag) into the video coding stream for transform coefficient values belonging to the sub-set of possible scan positions up to the last transform coefficient value belonging to a sub-set of positions, and After each significant flag of each non-zero transform coefficient value, depend on the respective transform coefficient value that is the last transform coefficient value belonging to the sub-set of the possible scan positions in the preset transform block that is non-zero or zero. By decoding the last-flag (last_significant_coeff_flag) And a video signal reconstruction from a picture quality-scalable video data stream. The method according to any one of claims 21 to 27, The preset scan order is to scan the transform coefficient values of the transform blocks such that transform coefficient values belonging to a higher scan position in the preset scan order correspond to higher spatial frequencies. Video signal reconstruction device from data stream. The method according to any one of claims 21 to 25, The parsing means The preset transform as well as the number of consecutive trailing transform coefficient values belonging to the sub-set of the possible scan positions within the preset transform block and having an absolute value of one of the nonzero transform coefficient values By decoding significant information belonging to the sub-set of the possible scan positions in the block and specifying a number of nonzero transform coefficient values; By decoding the consecutive trailing transform coefficient values and signals and signals of non-zero residual transform coefficient values belonging to the sub-set of the possible scan positions in the preset transform block; Transform coefficient values that are zero from the first transform coefficient values belonging to the sub-set of possible scan positions to the last transform coefficient values that belong to the sub-set of the possible scan positions in the preset transform block and that are nonzero By decrypting the whole; By decoding the number of consecutive transform coefficient values belonging to the sub-set of the possible scan positions in the preset transform block and immediately preceding any of the number of nonzero transform coefficient values in the inverse scan order, And reconstruct the successive portion. 23. A video signal reconstruction apparatus from a picture quality-scalable video data stream. The method according to any one of claims 21 to 29, The reconstruction means uses a motion-prediction based motion information and obtains motion-prediction residuals by inverse transformation of transform coefficient values into the block scheme of the transform blocks while obtaining the motion-prediction results to motion-prediction residuals. And reconstruct the video signal by combining the video signal reconstruction apparatus from a picture quality-scalable video data stream. The method of claim 20, The parsing means 402 is Each sub-data stream may comprise an indication for the respective picture quality layer to indicate presence or absence of motion information; and The sub-data stream of the first layer of the picture quality layers includes the motion information and has an indication indicating the presence of motion information, or the indication in the sub-data stream of the first picture quality layer includes the motion information. Indicate the motion information non-existence as part of the quality-scalable video data stream in addition to the sub-data streams, and The sub-data stream (s) of the other picture quality layer (s) may comprise an indication indicating the absence of the motion information, or may include inadvertent motion information with an indication indicating the presence of the motion information. So that And a video signal reconstruction device from a picture quality-scalable video data stream. The method according to claim 31, The parsing means is equal to the high definition motion information or refinement information which allows reconstruction of the high definition motion information based on the low quality motion information. A quality-scalable video data stream such that a sub-data stream can have an indication indicative of the presence of motion information, and wherein a portion of the quality-scalable video data stream is also formed to include the low quality motion information. Video signal reconstruction device from the user. The method of claim 31 or 32, And the parsing means is configured such that the motion information and the indication are related to a macroblock of the image. The method according to any one of claims 21 to 33, And said parsing means is configured to parse each sub-data stream separately from said other sub-data stream (s)-for parsing results-independently. . The method of claim 34, wherein The building means is configured to combine the transform coefficient values and the respective transform coefficient information independently of the other sub-data stream (s) from the picture quality-scalable video data stream. Video signal reconstruction device. The method of claim 35, wherein The layer order is defined between the quality layers, and The sub-data stream of the first quality layer in the layer order to combine the respective transform coefficient information with the transform coefficient values independent of the sub-data stream (s) of the subsequent quality layer (s). On the other hand, the sub-data stream (s) of the subsequent quality layers in layer order are simply combined with the sub-data streams of the quality layer (s) preceding the respective quality layers to provide the transform coefficient values and the respective values. To combine the transform coefficient values, Wherein said building means combines said transform coefficient values and said transform coefficient information of each of said image quality layers using said image quality layer (s) preceding said respective image quality layer and said sub-data streams of said each image quality layer. And reconstruct a video signal from a picture quality-scalable video data stream. The method of claim 35 or 36, The parsing means is formed such that the transform coefficient information of at least one of the image quality layers includes one distribution value relating to one transform coefficient value, And said building means is configured to sum said distribution values relating to said one transform coefficient value to obtain said one transform coefficient value. In each transform block, for each possible scan position, at least one of the transform coefficient values in each transform block has the preset scan positions 154, 156, 164, having the possible scan positions to belong to each of the possible scan positions. 166 is defined between transform coefficient values in the transform block, the video signal using block-wise transform to obtain transform blocks 146, 148 of transform coefficient values for the image 140 of the video signal. (18) encoding; And Transform coefficient information for transform coefficient values belonging to the sub-set of the possible scan positions such that the sub-set of each quality layer includes at least one possible scan position not covered by the other multiple sub-sets of quality layers. And forming, for each of the plurality of quality layers, a video sub-data stream (30; 28, 30) comprising scan area information indicative of the sub-set of possible scan positions; Quality-scaleable video data stream generation method comprising a. A method for reconstructing a video signal from a picture quality-scalable video data stream 36 comprising a video sub-data stream 30 (28, 30) for each of a plurality of picture quality layers, In each transform block, for each possible scan position, at least one of the transform coefficient values in each transform block has a preset scan order with the possible scan positions so that it belongs to each possible scan position. Determined for each image quality layer, to obtain, for each image quality layer, transform coefficient information and scan region information for transform coefficient values of different transform blocks, and the scan region information indicating a subset of the possible scan positions, wherein Parsing video sub-data streams of the plurality of quality layers; Constructing the transform block by using the scan region information, for each of the image quality layers, combining the transform coefficient values of the respective transform blocks from the transform coefficient information into a subset of the possible scan positions. ; And Reconstruction means (438) for reconstructing the image of the video signal by inverse transformation of the transform blocks; And a video signal reconstruction method from an image quality-scalable video data stream. In a picture quality-scalable video data stream enabling reconstruction of a video signal, Including transform coefficient information and scan area information on transform coefficient values of other transform blocks, In each transform block, for each possible scan position, a preset scan order having the scan positions possible above so that at least one of the transform coefficient values in each transform block belongs to each possible scan position is determined by the transform. Is defined between the transform coefficient values in blocks, and the possible scan region information indicates a sub-set of the possible scan positions, Wherein said transform coefficient information is related to transform coefficient values belonging to said sub-set of possible scan positions. Computer-program having a program code for performing the method according to claim 38 or 39 when operating on a computer.

Images